CC* Integration-Small: Enhancing Data Transfers by Enabling Programmability and Closed-loop Control in a Non-programmable Science DMZ

University Of South Carolina At Columbia, Columbia SC

Investigators

Abstract

Programmable data plane (PDP) switches have recently attracted significant attention from the communications industry. They are network devices that provide unprecedented visibility of events occurring in data networks and enable engineers to write software applications using the P4 programming language. The applications leverage the visibility and performance capacity of PDP switches, where applications can run orders of magnitude faster than those running on general-purpose computers. This project will deploy a "self-driving" network – referred to as Science DMZ – at the University of South Carolina (USC), using P4 applications running on PDP switches. The enhanced network will support science and engineering projects that foster the progress of science. Specifically, it will enable faculty members, researchers, and students to transfer big science data more efficiently within USC's campus and with external collaborators such as U.S. Department of Energy's laboratories (e.g., Savannah River, Argonne), national computing centers (e.g., San Diego Supercomputer Center, National Energy Research Scientific Computing Center), and international organizations (e.g., European Organization for Nuclear Research – CERN). The project has two objectives: 1) Develop a self-driving Science DMZ, capable of enhancing performance and fairness. Traffic will be monitored by PDP switches operating at terabits per second (Tbps) rate, and functions commonly executed on general-purpose CPUs will be offloaded to the PDP switches. The scheme will leverage the granular telemetry and the processing speed of the switches to automate the configuration of the Science DMZ and address challenges observed by USC and the community, such as traffic policing and optimal buffer sizing. Additionally, the project will extend perfSONAR to passively incorporate the telemetry generated by PDP switches, introduce real-time reporting on a per-flow basis, and add new metrics such as queue occupancy and packet interarrival time (i.e., per-packet visibility). 2) Disseminate the P4 technology by developing open-source P4 virtual lab libraries. The libraries will run on USC's cloud platform, referred to as the Academic Cloud, and on FABRIC. The Academic Cloud is used by several colleges and universities, researchers, and the cyberinfrastructure (CI) community for academic courses, workshops, and self-paced training. The libraries on FABRIC will use JupyterHub notebooks, incorporate the learner's customized topology, and embed P4 code alongside explanatory documentation, videos, and other visual aids. Finally, the project will support four PhD students and ten undergraduate students who will deploy the enhanced network under the guidance of the Principal Investigators and CI engineers. This project is jointly funded by the Office of Advanced Cyberinfrastructure, the Established Program to Stimulate Competitive Research (EPSCoR), and the Division of Computer and Network Systems. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.